In this preprint, Viki Brunner shows how, using the large CRyPTIC dataset, she can recapitulate the result that susceptible M. tuberculosis samples grow faster than samples that are resistant to rifampicin (and do not have any mutation that could compensate for that effect).
Using the Fisher’s exact test, she is able to confidently identify 51 putative compensatory mutations, having corrected for linkage disequilibrium, partly by insisting each mutation is homoplastic. Twelve of these hits have not previously been described.
There is then a very interesting story about how it appears at first glance that RIF-resistant samples with compensatory mutations grow better not just than those without, but also better than susceptible samples. When you break this down by lineage, it appears confounded with Lineage 2 and also with some clades in our dataset. Our data suggest there is something interesting here, but more detailed experimental work will likely be needed to disentangle exactly what is going on.